Abstract

Data streams have gained considerable attention in data analysis and data mining communities because of the emergence of a new classes of applications, such as monitoring, supply chain execution, sensor networks, oilfield and pipeline operations, financial marketing and health data industries. Telecommunication advancements have provided us with easy access to stream data produced by various applications. Data in streams differ from static data stored in data warehouses or database. Data streams are continuous, arrive at high-speeds and change through time. Traditional data mining algorithms assume presence of data in conventional storage means where data mining is performed centrally with the luxury of accessing the data multiple times, using powerful processors, providing offline output with no time constraints. Such algorithms are not suitable for dynamic data streams. Stream data needs to be mined promptly as it might not be feasible to store such volume of data. In addition, streams reflect live status of the environment generating it, so prompt analysis may provide early detection of faults, delays, performance measurements, trend analysis and other diagnostics. This thesis focuses on developing a data stream association rule mining algorithm among co-occurring events. The proposed algorithm mines association rules over data streams incrementally in a centralized setting. We are interested in association rules that meet a provided minimum confidence threshold and have a lift value greater than 1. We refer to such association rules as strong rules. Experiments on several datasets demonstrate that the proposed algorithms is efficient and effective in extracting association rules from data streams, thus having a faster processing time and better memory management.